San Luis Wind Sensor is wacky on directions

[windfind]

Should have mentioned that the Free membership only gets you low resolution models. They are OK for seeing the big picture and if the Bay Area will be windy. But they can even to begin to resolve topography like the San Bruno Gap, Hwy. 92 gap or Pacheco Pass. So they miss the venturi like effect that is so important in Bay Area winds.

The WF-WRF 1km model which we now run on the cloud is the best available on a regular basis for the Bay Area. For the America's Cup we ran a model at 250 meters. But only for the race course area. Very expensive! If we could only get about 20K more windsurfers and kiters we could have awesome forecasts.

Mike Godsey
iwindsurf.com/ikitesurf.com
Weatheflow.com

[ctuna]

I wonder what the wind farms around the area of San Luis and Rio
use for Modeling and if that is available to the Public. Is there any articles
that explainn the various models for wind prodictions?
I like to look at the direction the wind generators(windmills?) are pointing and there
general speed when driving down 152 to get a reference point for those
parameters.

[windfind]

Hi ctuna,

In parts of the USA wind farms contract with us for wind information. And they have hired several of our forecasters to do the wind forecasts for them. The wind farms have extensive sensor networks on their own turbines but they keep their data and modeling systems proprietary.

That said... the turbines have a much broader acceptable wind range than out kites and sails and on the west coast they are usually on hills or broad wind gaps so they do not need as high as resolution models as our customers do.

Weather modeling is extremely complex and many of the worlds fastest supercomputers are dedicated to running those models along with simulating things like fusion etc. The only reason a tiny company like Weatherflow can offer high resolution models is that the supercomputers do all the heavy lifting for us and then we can make high resolution versions of that data affordable by only covering extremely tiny portions of the world. e.g. Bay Area.

This link gives a simplified overview of how models work and links for more digging:

https://en.wikipedia.org/wiki/Numerical_weather_prediction

Mike
iwindsurf.com/ikitesurf.com
Weatheflow.com

[jcscott]

Hi Mike:
Thanks for the summary of how our winds are forecast. Having done (much simpler) modeling and simulation myself I understand how difficult and compute intensive it can be to get accurate estimates/predictions at short length scales and frequent time intervals.

Today, I have two questions.
1. Were you able to run the wind-tunnel model to estimate the effects of reservoir water level and other topography on the San Luis sensor?

2. Have you considered some of the recent tools developed by the machine learning community? Rather than trying to solve the physical model for a complex problem, they use deep neural networks to "learn" 10^n parameters of the model using training data, then use the trained network to generate predicted solutions based on current and recent observations.

For the San Luis site, the training might require historical data from all the stations in the immediate vicinity, plus a few sites in the Gilroy, Hollister area, and Los Banos and a few others in the Central Valley. Temperature, pressure humidity, wind-speed and direction are probably available at all. Cloud-cover and/or insolation (where is the marine layer?) would also be helpful. So there might be ~6 measurements, at about a dozen sites, with say reading taken every 15 or 30 minutes. Given that weather systems typically move at about 30 mph (plus or minus) training the network with this dataset would allow a few hours prediction. Add lower resolution measurements over a wider area, and it might be possible to forecast accurately days in advance.

Campbell